I lift this function out of message_store to
break some dependencies, and it's also more
consistent with the rest of the codebase:
alert_words.process_message
pm_conversations.process_message
recent_topics.process_messages
recent_senders.process_message_for_senders
We can do further cleanup to make these names
consistent (and possibly have them all work in
bulk), but that's out of the scope of the current PR.
We move the message_store.add_message_metadata function
(and all its dependencies) into a new module called
message_helper and rename the function to process_new_message.
(It does a bit more than adding message metadata, such
as updating our message store.)
We also have a "protected" interface now in message_store,
so that message_helper can access the message store:
update_message_cache
get_cached_message
Because update_message_cache is identical to
the former create_mock_message, we just renamed it
in the tests.
Most callers should use these functions:
message_helper.process_new_message (setting)
message_store.get (getting)
It's slightly annoying that the setter interface
is in a different module than the getter interface,
but that's how you break a bunch of dependencies.
We also extract the tiny message_user_ids class:
user_ids()
add_user_ids()
All the code moves here are pretty trivial, and
the code that was moved maintains 100% line
coverage.
The module name `message_helper` is not ideal, but it's a single
function and it'll save time to just do the topology change now and
leave thinking through the right name to later.
Before this we did not have remove event in server_events_dispatch.js
for the user group delete event even though server had. This was
leading to blueslip errors. Extracted the logic which was used in
success() of channel.del for user_groups into the remove case in
server_events_dispatch. Also removed the redundant reload call as
we already do that in server events.
This is a mostly verbatim extraction.
I re-phrased one line of code to work around a lint
false alarm. (Look for `preamble` in the diff.)
There are about 8 lines missing coverage here, so
the new module might be a good candidate to get
100% line coverage on.
Before this change, you would need to remove 74
edges from our dependency graph to make it
acyclic. Now it's 72.
This change does not impact the overall complexity
of our dependency graph (at least in terms of the
number of edges that we would need to remove to get
a tree), but it does clarify the picture a bit.
* `op` (operation) field, added in f6fb88549f, was never intended for
`custom_profile_fields` event. This commit removes the `op` as it doesn't
have any use in the code.
* As a part of cleanup, this also eliminates the schema check warnings
for `custom_profile_fields` event, mentioned in #17568.
When a user entered an invalid character (whitespace or characters not
present in a name), the cleaned-up array, and hence the query,
would be empty which resulted in an error.
Fixes#17542
Extend our markdown system to support mentioning of users
by id also. Following these changes, it would be possible
to mention users with @**|user_id** and silently mention
using @_**|user_id**.
Main intention for extending the mention syntax is to make
it convenient for bots to mention a users using their ids. It
is to be noted that previous syntax are also supported.
Documentation tweaked by tabbott for better readability.
The changes were tested manually in development server, and also
by adding some new backend and frontend tests.
Fixes: #17487.
These were introduced in ff9a929d7a
with no explanation of why they were necessary.
Generally you only render a few things, and it's
important that they're up to date.
We weren't doing a good job of invalidating the cache.
Eliminating the cache will fix bugs (like presence circles
being out of date) and break some dependencies.
I removed some very fragile test code that was relying
on invalid values taken out of the cache. (We now have
less line coverage, but if we want to test our rendering,
there are much cleaner ways to do it.)
As part of testing this, I renamed Hamlet to "aaron", so
that there are two aarons, and then I logged on as Iago
to see the "secondary" code in action that shows their
emails to distinguish them.
This add the schema checker, openapi schema, and also a test for
realm/deactivated event.
With several block comments by tabbott explaining the logic behind our
behavior here.
Part of #17568.
This adds support for unstarring all (starred)
messages from a particular topic, from the topic
popover.
The earlier implementation of this in #16898
was reverted in bc55aa6a01 (#17429)
because it had two problems-
1. The crash reported in bc55aa6a01
was due to message_store returning undefined. This happens
when the message itself hasn't been fetched from the server
yet, but we know that the message is starred from the ids
in `page_params` in `starred_messages.js`.
This commit handles this case explicitly.
Note that, we simply ignore those messages which
we haven't fetched, and because of this, it may
happen that we don't unstar some messages from that
topic. The correct implementation for this would
be to ask the backend for starred IDs in a topic.
2. The earlier implementation actually unstarred **all**
messages. This was because it grabbed the topic and stream_id
from the topic popover `data` attributes, after the topic
popover had been closed. This passed `undefined`, which
the function then interpreted as an action to unstar all
messages.
With this commit, we use the confirm_dialog widget,
which eliminates the need to store this data in the DOM.
This is mostly a refactoring to break the unnecessary
dependency of bot_data on settings_bots.
This is a bit more than a refactoring, as I remove all
the debounced calls to render bots during the
initialization of bot_data. (The debouncing probably
meant we only rendered once, but it was still needless
work.)
We don't need to explicitly render bots during
bot_data.initialize(), which you can verify by loading
"#settings/your-bots" as the home page. It was just an
artifact of how add() was implemented.
Note that for the **admin** screen, we did not and
still do not do live updates for add/remove; we only do
it for updates. Fixing that is out of the scope of this
change. The code that was moved here affects
**personal** bot settings.
Note that the debounce code is quite fragile. See my
code comment that explains it. I don't have time to go
down the rabbit hole of a deep fix here. The puppeteer
tests would fail without the debounce, even though I
was able to eliminate the debounce in an earlier
version of this fix and see good results during manual
testing. (My testing may have just been on the "lucky"
side of the race.) I created #17743 to address this
problem.
This mainly extracts a new module called
browser_history. It has much fewer dependencies
than hashchange.js, so any modules that just
need the smaller API from browser_history now
have fewer transitive dependencies.
Here are some details:
* Move is_overlay_hash to hash_util.
* Rename hashchange.update_browser_history to
brower_history.update
* Move go_to_location verbatim.
* Remove unused argument for exit_overlay.
* Introduce helper functions:
* old_hash()
* set_hash_before_overlay()
* save_old_hash()
We now have 100% line coverage on the extracted
code.
The only caller for this function was settings_config,
so we put it there.
For the stream_edit test we no longer mock the function.
(The reason we mocked the function was more about avoiding
the heavy settings_notifications import than the function
itself.) This gives some incidental coverage, but then I
also add some more real coverage on it.
We extract compose_fade_users and compose_fade_helper.
This is a pretty verbatim extraction of code, apart from adding a few
exports and changing the callers.
This change makes the buddy_data module no longer sit "above" these
files in the dependency graph (at least not via compose_fade):
* jquery
* lodash (not a big deal)
* compose_state
* floating_recipient_bar
* message_viewport
* rows
The new moules have dependencies that buddy_data already
had directly for other reasons:
* people
* util
And then buddy_data still depends on stream_data indirectly through
the compose-fade logic for stream_data. Even without compose-fade, it
would depend indirectly on stream_data via hash_util.
Note that we could have lifted the calls to compose_fade out of
buddy_data to move some dependencies around, but it's useful to have
buddy_data fully encapsulate what goes into the buddy list without
spreading responsibilities to things like activity.js and
buddy_list.js. We can now unit-test the logic at the level of
buddy_data, which is a lot easier than trying to do it via modules
that delegate drawing or do drawing (such as activity.js and
buddy_list.js).
Note that we still don't have 100% line coverage on the
compose_fade.js module, but all the code that we extracted now is
covered, mostly via buddy_data tests.
Previously, it was tedious to create actual message
objects in message_store for use in node tests.
This was mainly because, `add_message_metadata`
in message_store has many dependencies and
validation checks. Since it was difficult to create
actual message objects, many tests just mocked
the `message_store.get()` method to return the desired
message.
This commit adds a new helper method (`create_mock_message`)
to message_store, for use in node tests. This just stores
the object passed to it in the `stores_messages` map,
without any validation. We do not add any
default fields to the message object before saving
it from this helper, because doing so would decrease
the utility of this helper, and, if a test
depends on some field having a particular value,
then it would be better to just pass the field: value
pair from the test itself, for readability, rather
than relying on the helper to add the field for us.
This helper allows us to write deeper tests.
This commit also replaces some instances of mocking
`message_store.get()` to use this new helper method.
Fixes#17466
This commit will change encoding logic. Initial logic
was not encoding parenthesis, and this creates conflicts
with the markdown link format. To resolve this while encoding,
we're now replacing parenthesis with ".28" and ".29."
There is no need to change decoding logic because before
decoding any URL, we first convert all the “.” to “%.”
optimization: No need to replace parenthesis in popovers.js.
A few years ago I introduced the anti-pattern
of automatically calling success functions within
channel.get stubs. It's better to just capture
the success function and call it explicitly.
Also, we now have override(), so it's easy to
inline these types of things in a safe way.
These sigils will help make it easier to see that this is a special
Zulip syntax feature, not just something the other user typed, which
might help set the expectation that we're showing the time in the
user's timezone.
Tweaked by tabbott to improve variable/template naming.
We have generally gone away from using $(...)
initialization in modules that we test with
zjsunit, but there are a few remaining special
cases related to our billing and portico
codebases.
I remove an obsolete comment--we use get_streams()
for the `n` key now.
I also remove a guard statement from sort_groups()
that returned `undefined` for empty lists.
That guard statement would break this code:
const stream_groups = stream_sort.sort_groups(streams, get_search_term());
if (stream_groups.same_as_before && ...
The calling code prevents the situation anyway:
const streams = stream_data.subscribed_stream_ids();
if (streams.length === 0) {
return;
}
I modify the "no_subscribed_streams" test to test
the new behavior. (Even though stream_list currently
short-circuits the call here, that may change in the future.)
I also introduce the test() wrapper to explicitly clear
our data.
Use fully resolvable request paths because we need to be able to refer
to third party modules, and to increase uniformity and explicitness.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This introduces the make_stream_message()
helper to avoid all the strange
`messages[0] === message1` confusion.
We also clear data explicitly at the beginning
of the test.
It's still a messy test.
A bunch of tests were mutating the same message.
Now they just make their own copy.
I introduce the name "sample_message" for the common
message, but I remove the two bogus reactions, and
the "warning" test now just creates its own
test-specific data.
I considered using set_info({}, 0) to clear data,
but it has a lot of machinery that could lead
to accidental line coverage and/or extra test
complexity.
This commit adds a new class for typeahead items that do not need
a presence circle. It is changed to support addition of new items
that do not require presence circle.
There should not be any visual change due to this.
* move one test below setup
* add a test() wrapper to clear mutes
* split up a test
* use two different dates for java vs. gossip
* deterministically sort by date
All these changes make it so that the order of
the tests is a lot less fragile.
TextField is used to allow users to set long stream + topic narrow
names in the urls.
We currently restrict users to only set "all_messages" and
"recent_topics" as narrows.
This commit achieves 3 things:
* Removes recent topics as the default view which loads when
hash is empty.
* Loads default_view when hash is empty.
* Loads default_view on pressing escape key when it is unhandled by
other present UI elements.
NOTE: After this commit loading zulip with an empty hash will
automatically set hash to default_view. Ideally, we'd just display
the default view without a hash, but that involves extra complexity.
One exception is when user is trying to load an overlay directly,
i.e. zulip is loaded with an overlay hash. In this case,
we render recent topics is background irrespective of default_view.
We consider this last detail to be a bug not important enough to block
adding this setting.
We want to be careful about MIT code, as it leads
to more lax checks in the filter code.
And then we just use with_field in a couple places
and clear subscriptions.
For most functions that we were using __Rewire__ for,
it's better to just use the override helper, which
use __Rewire__ under the hood, but also resets
the reference at the end of run_tests.
Another nice thing about override() is that it reports
when you never actually needed the mock, and this
commit fixes the instances found here.
I didn't replace every call to __Rewire__. The
remaining ones fall under these categories:
* I looked for ") =>" in my code sweep,
so I missed stuff like "noop" helpers.
* Sometimes we directly update something
in a module that's not a function. This
is generally evil, and we should use setters.
* Some tests have setup() helpers or similar
that complicated this code sweep, so I
simply punted.
* Somes modules rely on intra-test leaks. We
should fix those, but I just punted for the
main code sweep.
This is a deceptively ugly diff. It makes
the actual code way more tidy.
I basically inlined some calls to mock_module
and put some statements in lexical order.
We now just use a module._load hook to inject
stubs into our code.
For conversion purposes I temporarily maintain
the API of rewiremock, apart from the enable/disable
pieces, but I will make a better wrapper in an
upcoming commit.
We can detect when rewiremock is called after
zrequire now, and I fix all the violations in
this commit, mostly by using override.
We can also detect when a mock is needlessly
created, and I fix all the violations in this
commit.
The one minor nuisance that this commit introduces
is that you can only stub out modules in the Zulip
source tree, which is now static/js. This should
not really be a problem--there are usually better
techniques to deal with third party depenencies.
In the prior commit I show a typical workaround,
which is to create a one-line wrapper in your
test code. It's often the case that you can simply
use override(), as well.
In passing I kill off `reset_modules`, and I
eliminated the second argument to zrequire,
which dates back to pre-es6 days.
I created this test in Jan 2019 with
6b7a4f8611.
At the time I created this we were still migrating modules
over from using the $(() => {...}) style of initialization,
so the test served a purpose as an easy smoke test that we
weren't doing something really stupid in
ui_init.initialize().
Now that that migration is complete, the test's warts start
to outweigh its benefits. Because it's just a smoke test,
we get credit for artificial line coverage where we're
exercising code without actually verifying anything other
than that it runs without crashing.
There are still possibilities for bugs within
ui_init.initialize() where we initialize modules in the
wrong sequence, but I am not confident that the node test
would catch them with any more likelihood (or clarity) than
either the puppeteer tests or czo bug reports.
And then on top of that, the zjquery setup in the node test
is hard to maintain. It's also hard to verify that some
of the setup isn't cruft.
If you want to see how much of a maintenance headache this
has been in recent times, just look at how many of the
recent commits have been related to things like es6
migrations or code cleanup sweeps.
We should just kill it.
"Alert Words" is one of Zulip's oldest settings UI elements, and as a
result is buggy. This commit converts it to use our standard
progressive-table-wrapper system used for settings tables, which has
the side effect of fixing a bug that mad ethe tables look pretty bad
if one adds a very long word.
Fixes#17172.
This commit addresses the problem of user's status visibility to
some extent. It adds presence circles, like we have in buddy_list to the
typeahead suggestions that are given for mentioning users in messages.
Tweaked by tabbott to adjust vertical alignment of group mentions as well.
Testing for the changes is done manually in the developement server,
and also by updating frontend tests to address these changes.
Fixes: #17138
Found by running the tests after
sed -i 's/\.with(/.toBeUsed().with(/g' frontend_tests/node_tests/*.js
Signed-off-by: Anders Kaseorg <anders@zulip.com>
Previously, exact matches could be pushed off the typeahead list in the
case where there were more prefix matches that happened to rank first,
which is confusing to the user: if an emoji, for instance, falls into
this category, it will never show up in typeahead, which is easy to
confuse with the emoji not existing.
This isn't a perfect fix — there are still cases where it's hard to find
emojis because the prefix-space is very crowded, but it does fix a
category of surprising and frustrating behaviour.
This doesn't come completely without downside - it means that the exact
match emoji will jump to the front of the list, which changes what is
currently conceptually a "filtering" operation to a "filtering and
sorting" operation, but it seems on the whole to be a more ideal
experience. This is particularly notable in the non-typeahead emoji
picker, which uses the same codepath, but this change seems somewhat
desirable even there, since it allows the user to type the name of an
emoji and press enter and have that emoji show up, without having to
visually confirm that they aren't inadvertently selecting a
prefix-matching emoji.
A better solution to this in the long term might be ordering emoji
results by shortest-first as a tiebreaker for alphabetical ordering,
since that should provide the same behaviour while keeping the mental
model as "filtering" (since the sort order won't change as the user
types), but this seems like a reasonable first pass, and changing to
shortest-first ordering after making this change won't break any muscle
memory for existing users.
We manually trigger a re-render of RT after a stream is muted
to update the list of topic in RT for the active filter.
This fixes the bug that RT doesn't update correctly
after a stream is muted.
If user is in private message narrow, we reduce height of stream
list to allow height for pm list in the left sidebar. We need
to recalculate it when moving out of pm narrow and moving in
rt narrow.
When idle, we try to backfill messages and in the end reselect
the closest message in the list, which can be a unread message
if present.
When recent topics is open, we can backfill messages; but
shouldn't select the message_id otherwise it will mark the
message as read if the message is unread while triggering
`message_selected.zulip`.
User can go from recent topics to stream / topic narrow via various
means, but all go through narrow.activate, hence we make sure all the
state changes we do in recent_topics.hide are actually applied when we
hide recent topics and go to another narrow.
This fixes the bug that narrowing from left sidebar to a stream
takes user to the top of the narrow.
We land user on the first row of the table instead of the search
box because here user can access hotkeys like `w`, `q`, `/`, etc,
which will not be directly available if user is focused in
recent topics search box.
For tests:
We set focus to search by default to avoid mocking a lot of
table html for getting the tests passing.
Previously the filter would be reset every time the page was
refreshed. This commit adds persistence via localstorage, the tests
follow the pattern used in tests for drafts.
Fixes: #15676.
Go to Recent Topics on "#", no hash and "#recent_topics".
Go to Recent Topics as the last destination for escape key.
Map `a` key to All messages and change its hash to
`#all_messages`.
Recent Topics is no longer an overlay now, but note that it is
also not a typical messages narrow. It can reside between
an overlay and a Filter in the sense that it is dispalyed as
a typical Filter narrow but has properties of an Overlay.
Compose box is not visible in this view as it will be confusing
to many users and hence compose shortcuts have also been disabled.
Keyboard shortcuts that apply on messages have also been disabled.
The remaining shortcuts that apply to a narrow are still accessible
here.
We now only assign target once, rather than
assigning it then overwriting it for the
not-sent-by-me use case.
I tried to extract a "selector" here but the linter
complained.
Splitting up the tests here ensures we don't
needlessly do extra work here. (In the prior
clumsy implementation, the second test that
I split out here would fail due to the lack
of us setting up the jquery stub.)
This warning was added in #6551. It’s not for any version of the
current Electron app, which we warn about on the server side with
DESKTOP_WARNING_VERSION, but rather some pre-Electron app so ancient I
don’t even know what it is. Apparently it communicated using the
window.bridge global, so eradicate that too.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
‘function’ and ‘=>’ are not equivalent because they bind ‘this’
differently. For these functions, the ‘function’ semantics are
intentional.
This reverts part of commit 1a241cef88
(#17388).
Signed-off-by: Anders Kaseorg <anders@zulip.com>
We now call $.clear_all_elements at the top
of run_test.
We have to exempt two modules from the new regime:
compose
settings_user_groups
Also, if modules do set_global("$", ...) we don't
try to call the non-existent function.
It's possible we'll want to move to something like
this, but we might want to clean up the two
sloppy_$ modules first:
// AVOID THIS:
// const $ = require("zjquery")
run_test("test widget", ({override, $}) => {
override(foo, "bar", ...);
$.create(...);
// do stuff
});
We weren't exercising this method in any
meaningful way during the tests, and when
do add coverage, we probably want to just
test it directly.
We also kill off stub_selector(), which was
never well-documented.
Callers can either explicitly pass in children,
stub out $(...)[0] as needed, or just
circumvent jQuery complications with override.
Note the reactions test was broken before,
since $(...)[0] was always returning the same
stub.
We no longer export make_zjquery().
We now instead have a singleton zjquery instance
that we attach to global.$ in index.js.
We call $.clear_all_elements() before each module.
(We will soon get even more aggressive about doing
it in run_test.)
Test functions can still override $ with set_global.
A good example of this is copy_and_paste using the
real jquery module.
We no longer exempt $ as a global variable, so
test modules that use the zjquery $ need to do:
const $ = require("../zjsunit/zjquery");
The "silent" option was kind of evil, as it had
$(...).find(...) passing back "self" instead of a stub.
Now we just use $(...).set_find_results(...) or
override(...) to simulate/bypass drawing code.
(It turns out hash_util didn't even need this option.)
The tooltips for the left panel of stream settings
have been broken since November 2018 due to my
commit 8f915da2ca.
The code prior to 2018 was restoring tooltips
right inside the loop where we were detaching
the row from the DOM to put it back into the
DOM at another place. And then I tried to
just add them in bulk, forgetting that I was
in the middle of all the DOM manipulation (and
hence my selector for the loop was a noop).
Also, I don't think we've ever had them for live
events that add streams. (I fixed that too.)
It's not clear to me that this code is actually
necessary, as we get hover help without
calling $(...).tooltip(...) properly.
This is probably why we didn't notice any
breakage when we merged my 2018 commit.
We just want to reset the scrollbar here, which
we still do via ui.reset_scrollbar.
You don't want to preserve scroll position if
you are filtering or re-sorting.
We have long had this annoying two-pass way of building the
DOM that I am trying to eliminate.
The function names that I introduce here describe the current
situation more accurately.
In passing I make it so that we only throttle redraws when
users are actually typing. Using a throttled redraw when
you click on the sort icons is at best unnecessary, and it
may actually aggravate double clicks.
This splits the one big `run_test` block into
add/edit/delete parts, for better clarity.
This also removes the `with_field` calls and
uses the override style instead, which is
preferred because it makes sure the stubbed
function is actually called.
This commit replaces `with_field` calls to
use the override style instead.
`override` is preferred since it makes sure
the stubbed function is actually called,
while `with_field` doesn't, which makes it
hard to spot dead code.
This commit replaces all `with_stub` calls and
explicitly calls `make_stub` instead.
The `with_stub` helper does not add much clarity
hence we now use scoped stub objects instead.
This de-indents some blocks where scoping isn't
required, for example when there is a single
stub object inside a `run_test` function.
With this change, we also need to explicitly
assert `num_calls`.
These are still kind of a mess.
The old code combined the worst of both worlds:
- we had one monolithic test
- we called the events multiple times,
verifying a different stub each time
Now I make the tests more granular.
We could actually re-combine the tests, but
in a nicer way, so that we just set
up multiple stubs and verify that all stubs
get correctly invoked.
There is also some code cleanup here--in dispatch_subs,
we don't stub stream_data, so it's easier to write
deeper tests that actually validate the data changes.
This prevents a bug where we interpret "2something"
as a modern slug instead of a legacy stream name.
The bug was probably somewhat unlikely to happen in
practice, since it only manifests if 2 is an actual
stream_id.
We still need to write to these globals with set_global because the
code being tested reads from them, but the tests themselves should
never need to read from them.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This commit makes it so that MessageListData
methods always attempt to filter muted messages.
We later, in a new function
(`messages_filtered_for_topic_mutes`)
check if `excludes_muted_topics` is true or not,
and skip the filtering work if it isn't.
This new function consistently returns a new list.
This refactor will later allow us to write clean
and concise code as part of mute users.
This commit also refactors the muting tests
for MessageListData, which were earlier
spread across two `run_test` functions.
These tests should remain organized,
since similar tests will be added as part of
user mutes in future commits.
Previously, the `muting_enabled` property of
MessageListData class was used to indicate whether
some messages in the message list need to be
filtered due to topic muting, depending on the
narrow. For example, we exclude messages belonging
to muted topics from stream narrows, but not from
search narrows.
The name `muting_enabled` is a bit confusing, and hence is
changed to `excludes_muted_topics`.
It is also important that the name be specific, since
a similar new property will be added for user mutes
in future commits.
I have a local branch with a hacked up version of
zjquery that lets you basically detect when zjquery
stubs are never actually invoked by real code.
There are some nuances to that kind of audit, so
I haven't pushed the auditing code, but these
are low hanging fruit.
I now just use inline the code to create stubs
for the line items in the markdown_content
container, and I don't add methods to the
zjquery stubs.
And then I use the new "children" feature in
zjquery's `$.create(sel, opts)` to set up
$(".markdown_content"), which means I don't
have to stub `each` any more.
It's actually pretty rare in our codebase to
call methods like `$(...).map` or `$(...).each`,
but we now support them better in zjquery.
You can pass a list of child elements now to
`$.create(...)`.
Fixes the sorting button labels in stream settings, which were
regressed by commit f8fbae4d8e (because
the HTML was not marked as being HTML).
Signed-off-by: Anders Kaseorg <anders@zulip.com>
This is a minor refactor in the muting test in
`message_list`.
The `unmuted_messages` function filters out messages only
considering topic mutes, and not stream mutes.
The test previously made it look like we were testing
stream muting, by stubbing the `is_topic_muted` on the
basis of `stream_id`.
This also replaces the stub and uses real data instead.
This is a prep commit, which renames some variables
and functions involved in topic muting to include
the word "topic" in them.
This is done to have clarity when similar code
will be added as a part of the mute-user in
future commits.
Replaced methods/functions of moment.js with date-fns library.
The motive was to replace it with a smaller frontend timezone library.
Date-fns ~ 11.51 kb
moment.js ~ 217.87 kb
Some of the format strings change because date-fns encodes them
differently from how moment did.
Fixes#16373.
It's not clear to me why this code was necessary,
and I assume it was either originally written
with a bit of misunderstanding of how zjquery
works or it became unnecessary with some refactoring
of the "real" code.
We move some of the data setup to the top of the file.
We also remove some get_sub() calls that aren't really
necessary now that peer_data and stream_data are more
independent.
The maybe_clear_subscribers() function was an artifact of
when we used to attach subscribers to the "sub" records in
stream_data.js. I think it was basically a refactoring
shim, and due to some other recent cleanup, it was only
used in test code.
We also change how we validate stream ids.
Going forward, peer_data just looks up stream_ids with the
normal stream_data API when it's trying to warn about
rogue stream_ids coming in. As I alluded to in an earlier
commit, some of the warning code here might be overly
defensive, but at least it's pretty self-contained.
We now use the same code in all places to
get the bucket of user_ids that correspond
to a stream, and we consistently treat
a stream as having zero subscribers, not
an undefined number of subscribers, in
the hypothetical case of us asking about
a stream that we're not tracking.
The behavior for untracked streams has
always been problematic, since if a
stream is untracked, all bets are off.
So now if we don't "track" the stream,
the subscriber count is zero. None of
our callers distinguish between undefined
and zero.
And we just consider the stream to be subscribed
by a user when add_subscriber is called,
even if we haven't been told by stream_data
to track the stream. (We also stop
returning true/false from add_subscriber,
since only test code was looking at it.)
We protect against the most likely source
of internal-to-the-frontend bugs by adding
the assert_number() call.
We generally have to assume that the server
is sending us sensible data at page load
time, or all bets are off.
And we have good protections in place
for unknown ids in our dispatch code
for peer_add/peer_remove events.
We also streamline some of the error handling code
by doing everything up front. This will prevent
scenarios where a single bad stream_id/user_id causes a
bunch of the same warnings in an inner loop.
This removes a bit of complexity. If a piece of
settings code needs to render a stream with
subscribers, it just asks for it.
We no longer have the brittle, action-at-a-distance
mechanism of mutating the subscriber count on to
the stream_data version of a sub.
Stream subs are pretty small, so making copies of
them is cheap, and the blueslip timings from the
previous commit can help confirm that.
There is some discussion of putting `subscriber_count`
on the Stream model, which may eventually get us
away from tracking it in `peer_data.js`, but we will
cross that bridge when we get there. See
https://github.com/zulip/zulip/issues/17101 for
more details.
The weekly stream traffic is a better tiebreaker
for stream typeaheads than subscriber count, as
it's more directly a measure of a stream's current
relevance.
Normally stream traffic and subscriber counts are
closely correlated, but a good example for me is
the #twitter feed on czo, which only has 80 subscribers,
but which gets more traffic than our #integrations
stream (with 16k subscribers). I would rather
see #twitter win the tiebreaker (if it even got
to the tiebreaker).
The main motivation behind this fix, though, is
to break our dependency on peer_data, which has
some upcoming changes that will introduce some
performance tradeoffs, and I want one less place
to audit.
Also, it will be easier long term to share this
code with mobile if we don't require mobile
to pull in our peer_data dependency. (The webapp
has different forces than mobile that dicate
our data structures.)
We use day_old calculated based on day instead of hours to
render last seen values. This fixes us incorrectly quoting
anything 24 - 48 hours ago as Yesterday and
incorrectly quoting `time` that are Yesterday
but < 24 hours ago in 'x hours ago' format.
We were adding `expanded` class to left-sidebar when searching
for streams even if the left-sidebar was not in the popover state.
This cased confusion with popovers.any_active returning true,
when actually it is not.
topic_generator previously included an entire lazy generator
combinator library that was used four times. These straightforward
equivalent loops might not be as fun but they are way simpler.
Signed-off-by: Anders Kaseorg <anders@zulip.com>
After this change all peer_data functions consistently
use stream_id rather than some "sub" object whose
data type is complicated by all sort of fields that
don't really concern how we track subscribers.
The goal here is to make all our peer_data functions
basically work in id space. Passing a full `sub`
to these functions is a legacy of when subscriber
info was attached to a full stream "sub" object,
but we don't care about anything sub-related
(color, description, name, etc.) when we are
dealing with subscriptions.
When callers pass in stream_id, you can be more
confident in a quick skim of the code that we're
not mutating anything in the "sub".
This de-clutters stream_data a bit. Since our
peer data is our biggest performance concern,
I want to contain any optimizations to a fairly
well-focused module.
The name `peer_data` is a bit of a compromise,
since we already have `subs.js` and we use
`sub` as a variable name for stream records
throughout our code, but it's consistent with
our event nomenclature (peer/add, peer/remove)
and it's short while still being fairly easy
to find with grep.
This sets us up to use better system-wide data structures
for tracking subscribers.
Basically, instead of storing subscriber data on the
"sub" objects in stream_data.js, we instead have a
parallel data structure called stream_subscribers.
We also have stream_create, stream_edit, and friends
use helper functions rather than accessing
sub.subscribers directly.
Steve Howell
tushar912
Anders Kaseorg
shanukun
Vishnu KS
Ganesh Pawar
Nishant Mittal
m-e-l-u-h-a-n
YashRE42
Abhijeet Prasad Bodas
100RABHpy
Nikhil Maske
Aman Agrawal
sahil839
Josh Gilley
Wesley Aptekar-Cassels
aryanshridhar
Rohitt Vashishtha